TL;DR: Over the past few days, two of the biggest answer engines were caught testing the same thing: moving their source citations to the bottom of the AI answer instead of inline. Google AI Mode now anchors you straight down to a citation card, and ChatGPT is experimenting with a separate overlay box for its links. Nothing about your ranking changed — but where a citation lives changes how many people ever click it. For agencies, this is a reminder that the goal was never “get cited.” It’s “get cited and be the answer someone still wants to open.”
What was spotted this week
On July 22, Search Engine Roundtable reported that both Google AI Mode and ChatGPT are testing citation cards placed at the bottom of the response, rather than woven into the text. The behavior was independently picked up by Optimixed the same day.
The two engines are doing it slightly differently:
- Google AI Mode — spotted by Sachin Patel — appears to anchor the user to the bottom of the answer, dropping them directly onto the citation card when they interact with certain results (the behavior showed up around free product listings).
- ChatGPT — spotted by Brodie Clark on his SERP Alert account — is testing a clickable overlay box for its sources. In his words, “ChatGPT seems to be experimenting with how it’s linking out to sources from its output.”
Both are tests, not permanent rollouts. Answer engines run dozens of these UI experiments at once, and plenty never ship. But when two competitors converge on the same idea in the same week, it’s worth reading the signal rather than the specific pixels.
Why citation placement is a business question, not a design detail
An AI answer has two audiences at once: the person reading it, and the sites being quoted to produce it. Those two interests pull in opposite directions. The engine wants to keep the reader inside the answer; the cited site wants the reader to leave and land on its page.
Citation placement is where that tension gets resolved. An inline citation sits right next to the claim it supports — it’s read in context, at the moment the reader is most curious about “says who?” A citation parked at the bottom, or behind an overlay, is further from that moment of curiosity and easier to skip. Same citation, different odds of a click.
This matters because for most publishers, the click is still the point. The zero-click reality of AI answers is well documented, and a citation you technically earned but that no one opens is a brand impression at best. Getting listed is the floor now, not the ceiling. (We dug into how hard citations already are to measure in how to actually measure AI visibility.)
What this doesn’t change
Before agencies over-react: the inputs to being cited haven’t moved. Engines still pull sources the same way — they favor content that directly and cleanly answers the question, that a crawler can read without executing JavaScript, and that states a claim in a self-contained, quotable block. None of that is affected by where the citation card lands in the UI.
So the playbook that earns the citation is unchanged:
- Answer the question in the first two sentences, then support it. That’s what gets lifted into an answer. We laid out the pattern in answer blocks.
- Be unambiguous about who you are and what you claim. Vague, hedged copy rarely gets quoted; specific, attributable statements do.
- Make the page trivially machine-readable — clean HTML, real headings, structured data — so the source card, wherever it renders, points at a page that loads fast and confirms the claim.
What this does change for how you advise clients
What moves is the conversion half of the story. If citations are drifting to the bottom of the answer across engines, then “we’re cited” is a weaker win to report on its own. Two adjustments:
- Report citations and clicks as separate metrics. Track whether a client is cited across the engines and whether those citations are actually sending traffic. Treating them as one number hides exactly the effect these tests create.
- Optimize the destination for a skeptical, late click. A reader who scrolls to a bottom citation and clicks anyway is high-intent but low-patience. The landing experience — does the page immediately confirm what the answer promised? — is now part of your AEO work, not a separate CRO project.
The through-line: as engines make it easier to stay inside the answer, the bar for earning the departure goes up. You want to be the source a reader deliberately seeks out, not just one of several cards they scroll past.
This is also why measurement can’t be a monthly screenshot anymore. Placement tests like this one can quietly change a client’s real-world results without changing a single ranking — which is the whole reason Preferium measures citations across all four answer engines rather than trusting a single surface’s UI to stay put.
Key takeaways
- Google AI Mode and ChatGPT were both spotted this week testing citations at the bottom of the answer — AI Mode anchoring to a citation card, ChatGPT via an overlay box (Search Engine Roundtable, Optimixed).
- These are UI tests, not confirmed rollouts — but two rivals converging in the same week is a signal about where the interface is heading.
- Citation placement changes clicks, not rankings. The inputs that earn a citation are unchanged; the odds a reader acts on it are not.
- The AEO fundamentals still hold: answer first, be quotable, stay crawlable and JavaScript-free.
- Split “cited” from “clicked” in client reporting, and optimize the landing page for a high-intent, low-patience visitor.
- Track citations continuously across engines — placement experiments can move results without touching your ranking.
